Last night OpenAI's API was throwing hardcoded text and their conversational prompt injection into responses. Error was resolved last night and we are looking for a back-up LLM for cases like this and adding additional protection around vendor errors.
Thanks so much for your patience.